As an anonymous researcher, writer, and programmer, Gwern Branwen has cultivated a significant following for his in-depth essays and data-driven analyses on a wide array of subjects. His work, primarily published on his website Gwern.net, is characterized by its meticulous research, intellectual honesty, and often contrarian viewpoints.

On AI and Technology

Much of Gwern's recent work has focused on the rapid advancements in artificial intelligence, particularly large language models. He is recognized for his early and detailed empirical theory of scaling laws in AI.

Quotes:

  1. "All of the pressures are to go bottom-up. From existing things, it's just much more palatable in every way to start at the bottom and replace there and work your way up, to eventually where you just have human executives overseeing a firm of AIs." [1]
  2. "The 10,000 foot view of intelligence, that I think the success of scaling points to, is that all intelligence is is search over Turing machines." [1]
  3. "AI models are already much more diverse cognitively than humans are. I think different LLMs think in very distinct ways that you can tell right away from a sample of them, right?" [2]
  4. "By writing, you are voting on the future of the Shoggoth using one of the few currencies it acknowledges: tokens it has to predict. If you aren't writing, you are abdicating the future or your role in it." [3]
  5. "If you take seriously the idea of getting AGI in a few years, you don't necessarily have to implement stuff and do it yourself. You can sketch out clearly what you want, and why it would be good and how to do it. And then just wait for the better AGI to come along and actually do it then." [3]
  6. "The AI kind of can't eat ice cream for you, right? It can't decide for you which kind of ice cream you like... only you can do that." [3]
  7. On a future with AI, he envisions a model with "a human CEO who does the vision thing, and then the AI Corporation kind of scurries around underneath them, doing, you know, the CEO's bidding, right?” [2]
  8. "You have one Steve Jobs-type at the helm, and then maybe a whole pyramid of AIs out there executing it and bringing him new proposals." [1]
  9. "There is no general Master Algorithm, and there's no special intelligence fluid. It's just a tremendous number of special cases that we learn, and then encode into our TPUs, right?” [2]
  10. "AI is a polymathic topic... there's no field or discipline that is not relevant to thinking about AI. Obviously you need computer science and hardware. But you also need things like primatology and understanding what changed between chimp and human brains, or the ultimate laws of physics that will constrain future AI civilizations." [1]

Learnings:

  1. Scaling Hypothesis: One of Gwern's most influential ideas is that intelligence, particularly in AI, is largely a result of scaling up computation, data, and model size. He was an early proponent of the idea that many breakthroughs in AI would come from quantitative improvements rather than purely qualitative algorithmic changes. This is a recurring theme in his essays on machine learning.
  2. The Bitter Lesson: Gwern frequently references Rich Sutton's "The Bitter Lesson," which posits that general-purpose methods that leverage computation are ultimately the most effective in AI research, outperforming approaches that rely on human-designed features and domain knowledge.
  3. AI-generated content and its future: Gwern has written extensively on the capabilities of models like GPT-3 for creative writing and other tasks, often demonstrating their surprising abilities and limitations through direct experimentation.
  4. The importance of data: A key learning from Gwern's work is the critical role of vast and diverse datasets in training powerful AI models. He has explored the contents of large datasets like the Pile and their influence on model behavior.
  5. AI safety and alignment: While optimistic about AI capabilities, Gwern also delves into the complexities of ensuring that advanced AI systems are aligned with human values, a field known as AI alignment.

On Self-Experimentation and Learning

Gwern is a proponent of N-of-1 or self-experimentation, using rigorous methods to test the effects of various interventions on himself.

Quotes:

  1. On his motivation for self-experimentation: "I'm a great believer in finding out for yourself."
    • Source: This ethos is evident throughout his self-experimentation essays, such as his work on darknet markets. Silk Road
  2. "The irony—that people want most the information they will learn from least—will not be lost on those familiar with signaling." [1]

Learnings:

  1. Spaced Repetition: Gwern is a long-time advocate for using spaced repetition systems (SRS) like Anki and Mnemosyne to efficiently memorize and retain information. He has written detailed guides and analyses of its effectiveness.
  2. Nootropics and Self-Blinding: Gwern has conducted numerous self-experiments on various nootropics (cognitive enhancers). A key learning from his methodology is the importance of self-blinding to mitigate placebo effects and obtain more reliable data.
  3. Dual N-Back Training: He has extensively researched and experimented with Dual N-Back, a working memory training task, concluding that while it improves performance on the task itself, the transfer effects to general intelligence are likely small or nonexistent.
  4. The Quantified Self: Gwern's work embodies the principles of the Quantified Self movement, which uses data tracking to gain insights into one's own body and mind. His detailed logs of experiments serve as a public example of this practice.
    • Source: His various self-experimentation reports, such as on Melatonin.
  5. Lithium in Drinking Water: In one of his notable investigations, Gwern analyzed the correlation between low-dose lithium in public drinking water and various societal benefits, such as reduced suicide rates.

On Psychology and Society

Gwern's essays often touch upon aspects of human psychology, societal trends, and the dynamics of online communities.

Quotes:

  1. "The most underrated benefit of anonymity is that people don't project onto you as much. They can't slot you into any particular niche or identity and write you off in advance. They have to at least read you a little bit to even begin to dismiss you." [4]
  2. "It is a high accomplishment indeed to factor out a bit of functionality into a library and make every possible user actually use it." [1]
  3. "One should be able to draw upon the wisdom of others." [1]

Learnings:

  1. The Melancholy of Subculture: This essay explores the life cycle of online subcultures, from their vibrant and creative beginnings to their eventual decline as they become more mainstream and commercialized.
  2. Are Sunk Costs Fallacies? Gwern challenges the conventional wisdom that succumbing to sunk costs is always irrational, arguing that in some real-world scenarios, it can be a rational heuristic.
  3. Terrorism is Not About Terror: He argues that the primary goal of terrorism is often not to instill fear, but to provoke a disproportionate response from the state, thereby radicalizing a broader population and achieving political objectives.
  4. The Collecting Mindset: An exploration of the psychology behind collecting, from stamps and coins to more esoteric items, and how it differs from hoarding or mere accumulation.
  5. Parasocial Relationships: Gwern has written about the nature of one-sided relationships with media figures and online personalities, a phenomenon that has become increasingly prevalent with the rise of social media.

On Writing and Research

As a prolific writer and researcher, Gwern has a distinct philosophy regarding his work and its presentation.

Quotes:

  1. On his writing ethos, learned from Wikipedia: "No code is so bad that it contains no good; the most valuable code is that used by other code; credit is less important than work; a steady stream of small trivial improvements is better than occasional massive edits." [1]
  2. "I'm trying to write more about what isn't recorded... things like preferences, and desires and evaluations, and judgments... things that an AI couldn't replace even in principle." [3]
  3. "I have derived a lot of benefit from people not being able to mail heroin to my home and call the police to SWAT me. But I always feel that the biggest benefit is just that you get a hearing at all. You don't get immediately written off by the context." [4]

Learnings:

  1. The Importance of Long-Form Content: Gwern's website is a testament to his belief in the value of detailed, long-form essays that thoroughly explore a topic, in contrast to the short, ephemeral content common on social media.
  2. Meticulous Sourcing and Linking: A hallmark of Gwern's writing is his extensive use of footnotes, links, and pop-up annotations, allowing readers to easily verify his sources and delve deeper into the topics discussed. This practice reflects a commitment to intellectual transparency.
  3. Anonymity as a Tool: Gwern's use of a pseudonym is a deliberate choice that he believes allows his work to be judged on its own merits, free from biases associated with personal identity.
  4. Open Source Information: Gwern makes his website and its contents freely available, embodying an open-source philosophy of knowledge sharing. The entire site is available on GitHub.
  5. The Value of Book Reviews: Gwern writes incredibly detailed book reviews, often approaching the length of his essays. This demonstrates a belief that critical engagement with the work of others is a valuable intellectual exercise.

Miscellaneous Insights

A selection of other interesting ideas from across his work.

Quotes:

  1. On the appeal of Dune: "They appealed, I think, to the sense that people have profound untapped powers that discipline can draw out." [5]
  2. "If [Nietzsche] doesn't make you wonder why you really believe what you do, then you are a clod." [5]
  3. "For many creatures, it just doesn't pay to be intelligent because that's not actually adaptive. There are better ways to solve the problem than a general purpose intelligence." [4]
  4. "There is nowhere in the brain where, if you hit it, you eliminate fluid intelligence. This doesn't exist." [4]

Learnings:

  1. The Economics of Darknet Markets: Gwern conducted pioneering research on the economics and operational security of darknet markets like the Silk Road, providing some of the first data-driven analyses of these platforms.
  2. Fashion Cycles: An analysis of the cyclical nature of fashion trends, applying mathematical models to understand how styles rise and fall in popularity over time.
  3. The Surprising Efficacy of Small Groups: Gwern has explored historical and contemporary examples of how small, dedicated groups can have an outsized impact on society.
  4. Literary Analysis of My Little Pony: Friendship is Magic: In a well-known and extensive essay, Gwern applies serious literary and philosophical analysis to the animated show, treating it as a complex cultural artifact.
  5. On the Nature of Beauty: An essay that delves into the concept of beauty, its objectivity, and how it has evolved, particularly in the context of aesthetics and art.
  6. The Value of Old Ideas: Gwern often resurrects and re-examines older, sometimes forgotten, scientific papers and ideas, demonstrating that valuable knowledge can be found outside of the current scientific consensus.
    • Source: This is a general theme, but his extensive archives of older papers are an example.
  7. Correlation vs. Causality: A recurring theme in his statistical analyses is a nuanced exploration of the relationship between correlation and causation, often going beyond the simple mantra that "correlation does not equal causation" to explore when and how causal inferences can be made.
  8. The End-to-End Principle in Machine Learning: Gwern advocates for the "end-to-end" principle in designing machine learning systems, where models learn directly from raw input to final output, minimizing manual feature engineering and intermediate steps.

Learn more:

  1. Gwern Branwen - How an Anonymous Researcher Predicted AI's Trajectory
  2. Takeaways From Gwern Branwen's Thoughts On AI Writing - Forbes
  3. Gwern - Anonymous Writer Who Predicted AI Trajectory on $12K/Year Salary - YouTube
  4. 101 Inspirational Quotes - Live Love Simple
  5. How to Use Learnings Correctly - Grammarist